Agents Need Headless Apps

Enterprise software vendors are rebuilding their products so artificial intelligence agents can act through APIs, not just click through dashboards. (blog.google) At Google Cloud Next on April 22, 2026, Chief Executive Thomas Kurian said the company was launching a Gemini Enterprise Agent Platform to “build, scale, govern, and optimize agents.” Google said nearly 75% of its cloud customers already use its artificial intelligence products, and 330 customers processed more than 1 trillion tokens each over the past 12 months. (cloud.google.com) Google also said its first-party models now process more than 16 billion tokens per minute through direct customer API use, up from 10 billion the prior quarter. That metric points to companies wiring models into software systems and workflows, not only into chat windows. (cloud.google.com) A “headless” app is software with its business logic exposed separately from its user interface, so another program can place an order, approve a refund, or update a record without opening the screen a person sees. In artificial intelligence systems, that means an agent can call stable interfaces instead of trying to imitate a human clicking around a browser. (developers.openai.com) That design is showing up across vendors. Salesforce said on April 15 that its new Headless 360 initiative exposes platform capabilities as APIs, Model Context Protocol tools, or command-line commands so agents can work directly with customer data and workflows. (ppc.land) The plumbing matters because one large agent often struggles with long, messy business processes. Microsoft wrote in October 2025 that enterprise systems are shifting toward “a network of specialized, atomic AI capabilities,” coordinated through workflow orchestration, routing rules, checkpoints, and shared state. (devblogs.microsoft.com) A second piece is the connector standard. The Model Context Protocol, introduced by Anthropic in November 2024, is an open standard for linking assistants to external tools and data sources, and OpenAI now documents support for remote Model Context Protocol servers in its Responses API. (anthropic.com, developers.openai.com) Vendors are also building governance into the stack because agents can take actions, not just draft text. Google’s Next announcements put “govern” alongside build and scale, and McKinsey wrote in March 2026 that moving from pilots to thousands of agents enterprise-wide requires policy, security, and oversight for autonomous decisions. (cloud.google.com, mckinsey.com) That shifts the work inside companies. Product teams have to expose core functions as dependable APIs, operations teams have to define permissions and audit trails, and developers have to design systems that agents can call safely over and over again. (blog.google, developers.openai.com) The pitch from cloud and software companies is no longer that an assistant will sit beside existing software. It is that the software itself has to be rebuilt so agents can operate it directly. (blog.google, ppc.land)